Agenda

• Why Modelling?
• Microsimulation Modelling
• Lifetable Modelling
• Summary

Why Modelling? (1/2)

• Simple representation of complex world/phenomena
• Experiments are not possible or feasible (e.g. randomised control trials)
• Explore alternative consequences using scenarios
• Trials are short (intermediate end-points), need for extrapolation Cholesterol levels (trial) to heart disease (model)
• Generalisable to other settings (age groups, country)
• Synthesizing data from various sources
• Informing decisions in the absence of hard data (recent pandemic)

Why Modelling? (2/2)

Courtesy: Randall Monroe, xkcd: <https://xkcd.com/1838>

Example - Transport PM 2.5 for Munich

How it differs from traditional models?

• Individual level (such as persons, households, vehicles, or firms)

• Autonomous entity - with limited learning/adaptability abilities

• Local interactions impact macro/aggregate levels

• Explorative that can be used for explanation or prediction

• Resource intensive - even with sample population

• Nuanced understanding

Ingredients

• Synthetic Population¹

• Social/physical space

• Rules of engagement with other actors/environment

Synthetic Population

• Census population

• Localized prediction model based on the most recent Census

• Surveys (like a travel or Physical Activity Survey)

Examples - using Bogota (1/2)

Examples - using Bogota (2/2)

Tutorial paper

Krijkamp, Eline M., et al. “Microsimulation modeling for health decision sciences using R: a tutorial.” Medical Decision Making 38.3 (2018): 400-422

Definition - Lifetable Modelling

• Two states model: alive and dead
• Outcomes: life years, and life expectancy.

Input/Outputs

• Inputs: Mortality rates
• Depicts: Life expectancy at different ages
• Period life tables: individuals exposed over hypothetical life course to mortality rates observed in one calendar year
• Projected mortality rates: Simulation relies on cohort life tables

Source of Input Datasets

• All-cause mortality rate: National Statistics offices/Global Burden of Disease Study

• All-cause mortality rate trends: National Statistics offices

• Population for each cohort (can also be 100,000 or similar figure)

Two-state lifetable model (1/3)

Two-state lifetable model (2/3)

Two-state lifetable model (3/3)

• Average life years without intervention: 6,362

• Average life years without intervention: 7,005

• Life years gained: 643 (7,005 - 6,362)

• Caveat

  • Probability of dying increases with age

  • Morbidity is captured by proportional multi-state lifetable models

Summary

• Microsimulation

  • Pros:

    • Large state spaces and captures diversity/heterogeneity on number of variables

    • May incorporate expert opinions

  • Cons:

    • Technologically challenging and resource intensive

    • (Yet) unsupported assumptions

• Lifetable Modelling

  • Pros

    • Relatively easy to set up

    • Comparatively low data requirements and easier to justify using Cost Benefit Analysis

  • Cons

    • Only mortality modelling

    • Typically based on annual rates

References

1. Veerman JL, Barendregt JJ,Mackenbach JP Quantitative health impact assessment: current practice and future directions Journal of Epidemiology & Community Health 2005;59:361-370.
2. Briggs, A.D.M., Wolstenholme, J., Blakely, T. et al. Choosing an epidemiological model structure for the economic evaluation of non-communicable disease public health interventions. Popul Health Metrics 14, 17 (2016). https://doi.org/10.1186/s12963-016-0085-1
3. Blakely, T., et al. (2020). Multistate lifetable modelling of preventive interventions: concepts, code and worked examples.
4. Barendregt, J. J., et al. (1998). “Coping with multiple morbidity in a life table.” Math Popul Stud 7(1): 29-49.
5. Moeckel, R., Spiekermann, K., & Wegener, M. (2003, May). Creating a synthetic population. In Proceedings of the 8th international conference on computers in urban planning and urban management (CUPUM) (pp. 1-18).

Access at https://t.ly/z5SO